There is a never ending demand for increased computer system performance. A common limiting factor in computer system performance is the path from the main central processing unit (CPU) to storage, or the I/O path. The CPU usually requires data from attached storage many times faster than the I/O path. Fibre Channel is a standard which addresses this I/O bandwidth limitation.
Fibre Channel is an American National Standards Institute (ANSI) set of standards which describes a high performance serial transmission protocol which supports higher level storage and networking protocols such as HIPPI, IPI, SCSI, IP, ATM, FDDI and others. Fibre Channel was created to merge the advantages of channel technology with network technology to create a new I/O interface which meets the requirements of both channel and network users. Channel technology is usually implemented by I/O systems in a closed, structured and predictable environment where network technology usually refers to an open, unstructured and unpredictable environment.
Advantages of Fibre Channel include the following. First, it achieves high performance, which is a critical in opening the bandwidth limitations of current computer to storage and computer to computer interfaces at speeds up to 1 gigabit per second or faster. Second, utilizing fiber optic technology, Fibre Channel can overcome traditional I/O channel distance limitations and interconnect devices over distances of 6 miles at gigabit speeds. Third, it is high level protocol independent, enabling Fibre Channel to transport a wide variety of protocols over the same media. Fourth, Fibre Channel uses fiber optic technology which has very low noise properties. Finally, cabling is simple in that Fibre Channel typically replaces bulky copper cables with small lightweight fiber optic cables.
Fibre Channel supports three different topologies, point-to-point, arbitrated loop and fabric, attached. The point-to-point topology attaches two devices directly. The arbitrated loop topology attaches devices in a loop. The fabric attached topology attaches a device directly to a fabric.
A Fibre Channel fabric is an entity which switches frames between connected devices. Fabric is a word which is synonymous with switch or router. The fabric must route the frame to the appropriate destination port or return a busy if the port is not available.
Because of the high link speeds, Fibre Channel fabrics face unique problems that are not present in current network switch design. Current network switches which support Ethernet, Fast Ethernet or Asynchronous Transfer Mode (ATM) protocols route incoming data at speeds up to ten to one hundred times slower than Fibre Channel fabrics. Current network switches also perform some incoming frame validation and network statistics collection. All these network switch features are more difficult to implement when the incoming frame rate is high, as in the case of Fibre Channel.
Route determination in network switches is usually performed by microprocessors. The requirement to route frames which are entering the fabric at speeds of up to one gigabit per second requires the fabric to route the frame in very little time. Routing depends not only on the incoming frame address but a host of other parameters and current state conditions as well. There are no currently available microprocessors which can in real time route sixteen lines of incoming frames with a link speed of 1 gigabit per second.
Frame validation creates another set of problems. In Fibre Channel fabrics frame validation must be performed at rates up to one hundred times faster than in Ethernet switches.
Statistics collection is also another function which must be performed in real time. Statistics collected are defined by the Fibre Channel fabric Management Information Base (MIB) and include the number of frames transmitted and received, the number of fabric rejects and fabric busies transmitted and received, etc. Gathering statistics for sixteen one gigabit per second ports creates new challenges.
Current fabric realizations use either fast microprocessors or digital signal processors to perform the route determination functions. Typically, processors are single instruction devices which serially decode the instructions and perform the specified function. Digital signal processors contain parallel functions and can perform several functions at one time. Still the problem exists to determine the route for many simultaneous incoming frames at one gigabit per second. Current fabric implementations perform routing on the order of tens of microseconds to hundreds of milliseconds. Ideally, routing should be accomplished in less than one microsecond.
Another problem with fabric realization is the support of the Arbitrated Loop topology. This topology has unique characteristics and requirements. Current fabric implementations do not support this topology.
Efficient support of both connection based classes of service (i.e., Class 1) and connectionless classes of service (i.e., Class 2 and 3) is also a challenge. A fabric must implement a different type of switch core to implement each class of service. Coordination between the different switch cores can be a burdensome task. Current fabric implementations support either a connection based or a connectionless switch core. This leads to inefficiencies, e.g., a connectionless switch core cannot switch Class 1 traffic if routes are not determined in frame time (i.e., less than one microsecond) and a connection switch core is very inefficient when routing Class 2 and Class 3 traffic.
Another problem with fabric realization is the interconnection or networking of fabrics. This is a problem due to the high speeds involved. Determining a network route is sometimes even more difficult than determining a local route. Destination addresses must be matched based not only on all bits matching but also matching a portion of the address. Route priorities should also be implemented to allow backup routes to a destination.